surajpaib/CT-FM-SegResNet

Instructions to load the pre-trained model weights

import torch
from monai.networks.nets import SegResNetDS

weights = torch.load("pretrained_segresnet.torch")

model = SegResNetDS(
    blocks_down=(1, 2, 2, 4, 4)
)

model.load_state_dict(weights, strict=False) # Set strict to False as we load only the encoder

# Dummy forward pass
tensor = torch.randn((2, 1, 32, 32, 32))

# Note that the input data needs to be in "SPL" format (OR z,y,x default numpy/torch format), 
# you can use Orientation transform in MONAI set with value "SPL". 
# Note: All subsequent transforms must be applied in (z,y,x) format. Eg patch size of [16, 32, 32] corresponds to 16 in z-axis
out = model(tensor)